Regarding the last few posts:

 * Raymond's implementation, while ugly, provides a completely orthogonal
   way to test compute numbits, useful in unit tests if nothing else.

 * Using x >> 1 in a reference implementation is perfectly reasonable.
   If the person using the reference implementation to produce a real
   C-based implementation doesn't understand the equivalence of x // 2
   and x >> 1, heaven help us.

Skip